Microsoft goes open source with one of its Bing algorithms
Microsoft's Space Partition Tree and Graph algorithm enables developers to apply vector search to traditional, audio and visual queries.
Microsoft has made its Space Partition Tree And Graph (SPTAG) algorithm, used in its own Bing search engine, available to all as an open-source GitHub project. The algorithm uses vector search and deep learning models to provide results based on search intent.
Understanding SPTAG. The algorithm is designed to allow users to search through billions of pieces of information, known as vectors, in milliseconds. In theory, this equates to more relevant results delivered more quickly.
Although not a new concept, this is made possible through vectorizing data, the process of assigning a numerical representation to a word, image pixel or other data point. By capturing the meaning of a piece of data in this fashion and applying deep learning models to associate it with other terms, Microsoft said it can begin to understand and represent search intent, which should mean results that match what the user actually wants (and not just the keywords they used).
In its blog post, Microsoft used the query, “How tall is the tower in Paris?” as an example. Even though the Eiffel Tower isn’t explicitly part of the query, Bing can still return a direct answer.
The implications. In making its algorithm available to the public, Microsoft continues its broader shift from being a closed ecosystem to one that’s more accessible and inviting, which may factor into the brand’s resurgence over the last few years. The fact that it was uploaded to Github, a Microsoft subsidiary, is also emblematic of its efforts to court the developer community.
Developers will be able to use Microsoft’s vector search technology to build their own search engines or help improve it by submitting updates. Outside of traditional search, the Bing team predicts that it’ll be used for enterprise or consumer-facing applications, such as identifying a spoken language via an audio snippet or determining an image’s content more quickly.
Why we should care. Getting closer to a searcher’s actual intent means we can learn about what they’re looking for and provide it; or if we can’t, we’ll waste fewer resources chasing uninvested users. Microsoft going open source with SPTAG is a good-faith gesture, but it also opens the door for developers to build upon the algorithm and potentially extend traditional, audio, and visual search in ways we’ve yet to imagine.
Opinions expressed in this article are those of the guest author and not necessarily Search Engine Land. Staff authors are listed here.